Closed nero19960329 closed 1 year ago
In addition, the reason why I adjusted the temperature of several agents to 0.5 and 0.7 is that in the first few experiments, I found that the system often got stuck in a few unsolvable problems, and the generated code was almost always wrong in the same way. Therefore, I wanted to try to increase some diversity in both the curriculum and action agents.
Hi, did you pull the latest version of Voyager? I fixed a typo about the chest selection in the latest commit. Sorry about the inconvenience.
Hi, did you pull the latest version of Voyager? I fixed a typo about the chest selection in the latest commit. Sorry about the inconvenience.
Haven't pull commit: https://github.com/MineDojo/Voyager/commit/220735c926a9664e9a139e0324ba3dcc8aadb593 yet, I'll pull the latest version and retry, thank you!
Before submitting an issue, make sure you read the FAQ.md
Briefly describe your issue
Deposit useless items into the chest at (xxx,xxx,xxx)
task successed once, then keep failing.It seems like the generated code failed at here: https://github.com/MineDojo/Voyager/blob/a6c1d9e58950d14798b09caaf299af015cf82724/voyager/control_primitives/useChest.js#L53
But I have no idea to fix this problem.
My
voyager_learn.py
:usage:
python voyager_learn.py --env-request-timeout 300 --max-iterations 300 --action-agent-temperature 0.7 --curriculum-agent-temperature 0.5 --curriculum-agent-qa-temperature 0.5
Please provide your python, nodejs, Minecraft, and Fabric versions here
[If applicable] Please provide the Minefalyer and Minecraft logs, you can find the log under
logs
folderDetails are at below:
[If applicable] Please provide the GPT conversations that are printed each round.
Details are at below: